upr client: Implement better error handling, retries and consider a different connection for control path operations

The upr client datapath (Event streaming) is a really heavy datapath and flow control is provided with the event stream protocol. We often need to request vbucket stats information from the ep-engine. It is noticed that sometimes during heavy workload, stats messages take more than 5 seconds and times out. We need to add better error handling for control related commands. Also consider the option of using a separate connection for stats.

[couchdb:error,2014-06-23T1:32:56.653,n_0@127.0.0.1:<0.21438.1>:couch_log:error:42]Uncaught error in HTTP request: {throw,
{timeout,
{gen_server,call,
[<0.1773.0>,

]}}}

Identified that stale=update_after latencies are affected by the slowness in obtaining vbucket stats